Cooperative coded matrix multiplication in secrecy-constrained vehicular networks

We propose a new class of polynomial codes, distributed structured PolyDot codes (DSPolyDot codes), for cooperative computation in vehicular edge computing networks, where computational tasks are abstracted by a matrix multiplication problem A⊺B, where A captures input data (features) and B the model parameters (weights), with entries over Fq, where the matrices are stored in separate source nodes, e.g., the central cloud network and the base station. In cooperative vehicular networks, source data must be processed across distributed vehicles and roadside units that may not be fully trusted, motivating the need for information-theoretic secrecy in terms of bounded information leakage at each cooperative node. Leveraging the algebraic structure of matrix multiplication, we design a coded computation framework that ensures information-theoretic secrecy with bounded information leakage at each cooperative node while naturally tolerating stragglers and achieving the same communication and computation efficiency as state-of-theart polynomial codes in the asymptotic regime. These make DSPolyDot codes well-suited for secrecy-constrained cooperative signal processing applications in vehicular wireless networks.